Define linear model for fractional values of `linear.years`

[rplzzz]

Right now when running samples we round our sample values of linear.years to the nearest integer. In the long run that is going to create some problems for doing inference with these samples. We can (relatively) easily define the model for fractional values by interpreting a fractional number of years as a weighted average of the two surrounding integer values. Thus, linear.years==3.4 is a 40% weighting of a 3-year model averaged with a 60% weighting of a 4-year model. This will allow us to avoid having to do inference with a mixture of discrete and continuous parameters.